Unable to replicate performance
#1
by
toretto14
- opened
Hi, I'm trying to replicate the training on common voice corpus since the WER is still converging in the plot you shared so i'm trying to see how far it can converge before training on some new datasets like MGB-2 but in best case the eval WER is not going below 0.5 for me, can you please share some more details of training script and hyper-parameters you used.
Thank you.
Hi, the script (which includes the default parameters' values) are in the model card: https://github.com/huggingface/transformers/blob/1c06240e1b3477728129bb58e7b6c7734bb5074e/examples/research_projects/wav2vec2/finetune_large_xlsr_53_arabic_speech_corpus.sh
elgeish
changed discussion status to
closed